DISCUSSION

6.1. AIRE ability to activate reporters without regulatory sequences is in line with its universal capacity to

augment gene expression

Our first finding in this study was that AIRE is able to activate Luc reporters without promoter sequences (Study I). This unusual ability of AIRE could be interpreted as the capacity to efficiently use short cryptic sequences for transcrip-tion activatranscrip-tion or to push pre-formed initiatranscrip-tion complexes outside gene promoters into productive elongation (Venters and Pugh, 2013). That could also explain described AIRE usage of alternative TSSs (Villaseñor et al., 2008). Though, in our total RNA-seq experiments we could not detect transcripts outside annotated gene starts, which could be due to their unstable nature (Davis and Ares, 2006;

Thompson and Parker, 2006). The promoter strength is decisive for AIRE-mediated activation. While reporters containing none or weak promoters, such as endogenous target gene, are activated by AIRE (Study I), strong promoters (SV40 and CMV) are not subjected to AIRE influence (Giraud et al., 2014).

Our current (Study I) and many previous works (Mathis and Benoist, 2009;

Peterson et al., 2008) suggest that AIRE does not have any sequence specificity for DNA binding, despite some reports claiming inclination for certain motifs (Kumar et al., 2001; Purohit et al., 2005; Ruan et al., 2007). Both genomic se-quences and reporter vectors could contain cryptic transcription initiation sites (Lemp et al., 2012; Vopálenský et al., 2008) due to the indiscriminate nature of eukaryotic transcription. Scanning for specific elements might not be useful as Inr similar dinucleotide combinations could occur very frequently. Transcription in the mammalian genome can be initiated at many loci, with productive tran-scription being limited to certain functional product containing genes (Kapranov et al., 2007). The biological significance of such pervasive transcription is not completely understood (Jacquier, 2009), but could explain why we observe reporter induction by AIRE with plasmids without any regulatory elements.

AIRE-induced promiscuous expression has been shown to have strong fea-tures of stochasticity in thymic epithelial cells (Guerau-de-Arellano et al., 2008;

Jiang et al., 2005; Venanzi et al., 2008; Villaseñor et al., 2008). However, euka-ryotic gene expression is not deterministic in general (Blake et al., 2003; Elowitz et al., 2002). Stochastic mRNA synthesis in genetically identical mammalian cell population was demonstrated by counting individual molecules of RNAs pro-duced from the reporter integrated in the genome of Chinese hamster ovary (CHO) cells (Raj et al., 2006). Accordingly, it has been proposed that mRNAs are synthesized in short but intensive transcription bursts with burst size regulated by promoter type and burst frequency by enhancer (Haberle and Stark, 2018).

Modulating transcription burst frequency through enhancer is more prevalent way to regulate the intensity of transcription (Fukaya et al., 2016). In light of this

knowledge, AIRE seems to be a unique factor that has capacity to induce stochastic pervasive background transcription potentially present in every cell.

The presence of intron can drastically increase the effectiveness of transcrip-tion, for example, beta globin intron intensifies RNA production hundredfold in plasmid reporter (Buchman and Berg, 1988). Alternative splicing and poly-adenylation are the ways to expand the repertoire of isoforms, and presentation of those different protein forms in the thymus can be crucial for avoiding auto-immunity (Klein et al., 2000). Considering the interaction of AIRE with splicing and RNA processing machinery (Abramson et al., 2010), we tested how the presence of introns and poly(A) sequences influences AIRE-mediated transcrip-tional activation. We found that AIRE can activate to similar extent intronless plasmids and reporters containing histone stem-loops instead of poly(A) signal.

In our RNA-seq experiments, we found that AIRE enhanced alternative splicing (Study III), but alternative exon usage was noted in a subset of genes different from differentially expressed genes. While transcription and splicing are tightly interconnected processes (Herzel et al., 2017; Kornblihtt et al., 2004) with splicing taking place as soon as nascent transcript emerges and exon inclusion is influenced by the speed of progressing Pol II (de la Mata et al., 2003), it is intriguing to observe such division into two separate gene sets.

6.2. AIRE binding to DNA

We characterized the DNA-binding ability of AIRE protein mediated by extended HSR/CARD+ domain (1143 aa). Within this domain, mutations that disrupt the integrity of HSR/CARD (1100 aa) and mutations of amino acids 113-114 out-side the HSR/CARD domain drastically reduced the binding (Study I). Although amino acids 113-114 in AIRE protein were earlier described to be responsible for nuclear transport (The Finnish-German APECED Consortium, 1997), their mutations do not interfere with AIRE nuclear localization (Ilmarinen et al., 2006).

In other proteins, DNA-binding domain and NLS signal overlap has been described, which could be evolutionarily beneficial, as DNA-binding proteins should be transported into and kept in the nucleus to exert their functions (Boulikas, 1994; Cokol et al., 2000).

Several studies have shown or suggested AIRE binding to DNA. The first report about AIRE binding to DNA was published in 2001, when PHD1 with potential leucine zipper was proposed to be responsible for AIRE dimerization and DNA interaction (Kumar et al., 2001). In this study, AIRE monomers were not able to bind oligonucleotides, but dimers and tetramers obtained by oxidation and refolding method, usually applied to isolate protein aggregates from bacterial inclusion bodies, showed preferential binding to GG in A/T rich environment (Kumar et al., 2001). Albeit the binding to DNA was firstly attributed to PHD zinc fingers, the function was later associated with SAND domain (Purohit et al., 2005). Though, as in a more recent work, the mutations introduced into the potential DNA-binding motif of SAND domain did not interrupt DNA binding, a

strong interaction of AIRE with DNA was alternatively proposed to be mediated via DNA-PK (Zumer et al., 2012). However, EMSA experiments shown here (Study I) and by others (Koh et al., 2008) found AIRE binding to DNA to be independent of DNA-PK or other proteins. Furthermore, EMSA experiments demonstrated AIRE interaction with both reconstituted mononucleosomes and the free DNA used to reconstitute them, but with the greater affinity towards naked DNA (Koh et al., 2008). Importantly, the removal of PHD1, PHD2 or SAND domains did not disrupt this binding (Koh et al., 2008), which is in line with our results demonstrating that AIRE DNA binding is mediated by extended HSR/CARD+ domain. Although initial transactivation experiments in 2000 were conducted tethering of AIRE fusion protein to reporter plasmids via Gal4-binding domain, the phenomenon that AIRE was able to activate the same plasmids even without Gal4 fragment was noticed (Björses et al., 2000; Pitkänen et al., 2000). While Gal4 tethering was still used in later experiments (Žumer et al., 2011) to achieve stronger activation (up to hundredfold) (Björses et al., 2000), our results confirm that AIRE is able to activate reporter genes efficiently (10-30-fold), even without fused additional DNA binding domains (Study I).

Several studies have demonstrated that AIRE target genes are characterized by low level of initial expression and repressive chromatin marks H3K27me3 and H3K9me3 (Handel et al., 2018; Sansom et al., 2014) as well as lack of active marker H3K4me3 (Koh et al., 2010; Org et al., 2008). In contrast, ChIP-on-ChIP experiments showed that the strongest AIRE binding at TSS of genes with high expression and high Pol II levels, however, the expression of these genes was not influenced by AIRE (Giraud et al., 2012). These ChIP-on-ChIP results could reflect the ubiquitous preferential binding of AIRE to accessible chromatin, which is characteristic to many TFs (Lambert et al., 2018). The interaction with accessible chromatin suggests that for the specific target gene activation, AIRE requires additional factors, which are present at genes with low expression level, or just indicates the inability of AIRE to enhance the expression of highly expressed genes. Considering the reported capacity of AIRE to release paused Pol II, the recirculation of transcription machinery on highly expressed genes could be already so effective, that AIRE might not be able to increase it. In our ChIP experiments, the full-length AIRE enrichment did not show differences between AIRE target and tested housekeeping genes (Study I). This could be because of differences in chromatin structure of HEK293 cells as compared with mTECs, altered modification pattern of overexpressed transfected AIRE versus endogenous AIRE, or relative insensitivity of qPCR-based analysis of ChIP material. On the other hand, ChIP with extended HSR/CARD+ domain demon-strated the highest binding to promoters of highly expressed genes and not near target genes (Study I), which indicates that previously observed AIRE prefe-rential binding to TSSs of highly expressed genes (Giraud et al., 2012) may also reflect the capacity of AIRE to directly bind to DNA via HSR/CARD+ domain.

AIRE ability to favor paused Pol II into productive elongation was demon-strated by applying specifically designed plasmids that differ in ability to recruit and load Pol II at promoters and thus to induce initiation and/or elongation (Oven

et al., 2007). This finding was supported by detecting on wide scale first exon skewing of Aire target genes with more transcripts produced from the first exon relative to the whole gene in Aire knock-out mice, thus demonstrating Pol II promoter-proximal pausing in the absence of Aire (Giraud et al., 2012). However, we could not detect increased Pol II CTD Ser2 modification accompanying the transition from initiation to elongation stage on AIRE target genes in genomic context (data not shown). Elevated levels of Pol II phosphorylated at Ser2, especially near 3’-end, were reported on plasmid target when AIRE protein was tethered by Gal4 (Žumer et al., 2011). We could detect high amount of AIRE protein bound to naked plasmid DNA devoid of nucleosomes and histones, and the ability of AIRE to activate reporters in these conditions (Study I). Interes-tingly, the addition of etoposide did not enhance the expression of plasmid reporters further (data not shown), indicating that additional DNA breaks created by inhibiting TOPs and thereby increased amount of chromatin-free DNA may own more importance in genomic context.

At first glance, AIRE deviates from traditional TFs, though, accumulating evi-dence of eukaryotic transcription suggests that simple model of DNA recognition motif dictating specific binding sites near the regulated genes may not be neces-sarily the case for the majority of eukaryotic TFs (Lambert et al., 2018).

6.3. AIRE-mediated transcriptional activation involves DNA damage repair machinery

In the current study, we identified AIRE interaction with DNA-PK, a protein mainly associated with DNA damage repair (Study II). The functional signi-ficance of AIRE interaction with DNA-PK has been intriguing question. Despite the extensive studies on DNA-PK role in DNA repair, the first reports about the DNA-PK protein described its involvement in transcription by phosphorylating SP1 (Jackson et al., 1990) and paused RNA Pol II (Dvir et al., 1992). We demon-strate here the ability of DNA-PK to phosphorylate AIRE at N-terminal amino acids T68 and S156 and show that mutations of these amino acids to alanines decrease the target gene activation (Study II). In subsequent work, inhibition of DNA-PK catalytic activity with NU7441 did not suppress AIRE-mediated tran-scriptional activation, invoking alternative hypothesis of DNA-PK requirement for AIRE binding to DNA (Zumer et al., 2012). Later, thorough screening of AIRE interacting proteins revealed several other proteins to be involved in DNA damage repair processes, including helicases, TOPs and ribosyltransferases to associate with AIRE (Abramson et al., 2010; Gaetani et al., 2012). Abramson et al. proposed that AIRE mimics the action of etoposide by stabilizing TOP2A-created breaks that activate DNA-PK and other DNA repair associated proteins.

Of note, this hypothesis was made because etoposide but no other DNA break-inducing compounds, for instance, H2O2, demonstrated the activating effect on AIRE target genes (Abramson et al., 2010).

To test the hypothesis about similarity of AIRE and etoposide effects, we used HEK293 cell line with inducible AIRE and studied the effects of etoposide, AIRE and their co-influence in these cells. In whole transcriptome analysis with doxycycline-inducible HEK293 cell line, we observed a strong activating effect of AIRE with 691 genes upregulated and only one downregulated (Study III).

Predominant transcription activation was described earlier by comparison mTECs of Aire-deficient and Aire wt mice, with a small fraction of genes down-regulated (Meredith et al., 2015; Sansom et al., 2014). More downdown-regulated genes can be explained by prolonged presence of AIRE in the cells, AIRE effect on mTEC differentiation or other pathways involved in mediating AIRE impact. In contrast, recent report demonstrated AIRE repressive effect at chromatin accessi-bility and transcriptional levels (Koh et al., 2018). Although we could observe chromatin compaction at sites distant from AIRE target genes (Study III Figure 6A, 6B), we detected undeniable upregulating effect of AIRE on its target genes in HEK293 cells.

We observed the enhancing effect of etoposide on AIRE target genes, in-creasing the expression level and expanding the repertoire of activated genes (Study III). We could not observe activation-augmenting effect of etoposide when assaying for plasmid reporter transcription in AIRE-positive cells (data not shown), implying that chromatin context is important for etoposide-mediated gene expression changes. Interestingly, we ourselves could not detect the syner-gistic effect of combining AIRE and etoposide when etoposide was applied before and AIRE expression induced after that (data not shown). This highlights the etoposide importance for AIRE-mediated activation in genomic landscape and suggests that DSB are not important for AIRE recruitment. As AIRE-dependent genes are embedded within relatively closed chromatin, it is possible that etoposide capacity to stabilize DSBs in this context is more influential on transcription rate than in case of plasmid where transcription machinery can process relatively easily as soon as recruited.

Besides TOP2 inhibitor etoposide, we observed similar activating effect of AIRE target genes when TOP1 inhibitor camptothecin was used while topoiso-merase inhibitors that block TOP1 or TOP2 catalytic activities entirely without DSB formation did not affect AIRE-induced gene expression activation (Study III). This indicates that stabilization of either SSBs or DSBs or the prolonged presence of TOP1 or TOP2 have positive effect on AIRE-mediated gene activa-tion. Interestingly, TOP1 was shown to be involved in initial complex assembly at superenhancers and enriched at H3K27ac rich distant sites, at the same time TOP2A was accumulated at TSS and involved in subsequent transcriptional events (Bansal et al., 2017). The same study found TOP inhibitors to have repressive effect on genes upregulated by AIRE (Bansal et al., 2017), which was seemingly in conflict with earlier findings of Abramson et al and ours (Study III;

Abramson et al., 2010). The differences between the studies could stem from the way etoposide was introduced  adding to the cell culture medium or injecting intraperitoneally into the mice for three days (Bansal et al., 2017).

To our knowledge, no whole transcriptome screening was performed with etoposide earlier, although individual targets are reported (Collins et al., 2001;

Tammaro et al., 2013). We observed a surprisingly large proportion of up-regulated genes when uninduced HEK293 cells were treated with only etoposide at low dose (2μM) (2452 genes activated (55,6 %) and 1961 downregulated (44,4

%)), unlikely to be attributed to secondary effects, such as stress response (Study III). Considering that etoposide induces DSB formation, such large number of activated genes was not expected. In RNA-seq experiments, we also detected a considerable set of genes, where alternative splicing was influenced by etoposide.

Etoposide influence on alternative splicing was demonstrated in proteomic analyses, where etoposide treatment of osteosarcoma cells evoked modifications of proteins involved in RNA metabolism (Beli et al., 2012), including dephospho-rylation of SRSF1 (Montecucco and Biamonti, 2013), leading to alternative splicing of genes involved in the choice between pro- and anti-apoptotic path-ways (Montecucco et al., 2015). Assessing thoroughly transcriptional and co-transcriptional effects of topoisomerase inhibitors could bring new knowledge and attenuate their usage in cancer therapies.

L28P mutation in HSR/CARD domain disrupts binding to plasmid DNA (Study I) as well as interaction with TOP2A (Study III). MNase treatment of lysates before IP does not exclude the participation of DNA or other proteins mediating AIRE-TOP2A interaction, as HSR/CARD domain is important for oligomerization and copious contacts with other proteins and DNA (Study I;

Pitkänen et al., 2000). The potential order of complex formation between AIRE, TOP2A and DNA is not defined, and initial event is not determined. However, in a previous report, the treatment with DNA intercalator ethidium bromide disrupted complex formation, suggesting that complexes are not pre-formed but depend on DNA presence (Bansal et al., 2017). In our experiments, we observed that MNase treatment, but not ethidium bromide, weakened AIRE─DNA-PK complex formation (Study II).

6.4. Stochastic nature of AIRE-mediated activation One of the first analysis of target gene expression at the single cell level revealed a contrasting pattern of casein locus regulation in mTECs and mammary gland epithelial cells (Derbinski et al., 2008). While tight co-expression of milk proteins was observed in mammary gland cells, mTECs did not comply with the same rules and rarely co-expressed the same set of genes in each individual cell. This observation emphasized again the difference between stochastic nature of pro-miscuous gene expression in mTECs and cell lineage-specific gene control. The tendency to accumulate in clusters could reflect the maturation program of mTECs, as gene clustering is also distinguishable in Aire-negative mTEChi cells, although the number of clusters and genes per cluster are reduced (Derbinski et al., 2005). We could observe clustering of AIRE target genes in HEK293 cells as well (Study III), indicating that the feature is not exclusively inherent for mTECs.

The propensity to form clusters could refer to AIRE ability to delineate transcrip-tional boundaries, occupying superenhancers (Bansal et al., 2017) or interacting with CTCF and other insulators. The formed cluster increases the probability of expression and predisposes genes for transcription induction; however, the final productive expression may be achieved by additional mechanisms with random selectivity.

Single-cell RNA-seq identified microclusters co-occuring in very few cells, suggesting these these cells developed from the common progenitor sharing same epigenetic signatures (Meredith et al., 2015). The identified clusters contained activated genes localized on different chromosomes, suggesting that inter-chromosomal contacts in mTECs are more prevalent, than cis-interactions (Mere-dith et al., 2015). Spatial organization could play critical role in gene expression (Dixon et al., 2016; Vermunt et al., 2019). AIRE forms distinct speckles inside the nuclei (Ramsey et al., 2002) and is shown to associate tightly with nuclear matrix resulting in non-random localization of AIRE-responsive genes (Tao et al., 2006). There are several possibilities to create various territories inside the nucleus, starting from every chromosome occupying certain territory (Cremer and Cremer, 2010) ending with diverse cis-regulatory elements including en-hancers, silencers, insulators, tethering elements that provide boundaries and scaffold for spatial architecture (Spitz and Furlong, 2012).

Despite differences in AIRE actions on plasmid target and in genomic context, the in vitro studies with transfected AIRE in different cell types have provided invaluable insights into AIRE functioning mechanisms. Although some results are conflicting and not all of them are readily translatable to in vivo situations, the pieces of AIRE transcriptional puzzle are coming together. Based on current knowledge and our three studies, we propose a following model for AIRE action as one option. First, when AIRE expression is activated, it localizes into the nucleus and binds both at active and inactive genes. Direct binding of extended HSR/CARD+ domain to DNA enables interaction with genes with less packed chromatin, and repressive chromatin marks consolidate binding via PHD1 to closed chromatin. AIRE binding is accompanied by recruitment of TOP proteins that cleave DNA and by activation of DNA-PK and other DNA repair proteins.

As a result, AIRE gets phosphorylated, chromatin more relaxed, and interaction with P-TEFb and BRD4 established via superenhancer contact, culminating in Pol II pause-release and productive transcription. Many other interacting proteins support these chromatin and spatial changes. While functionally the importance of AIRE binding to inactive chromatin is more obvious, the rationale of inter-action with active genes has yet to be explored. One option may be to increase

As a result, AIRE gets phosphorylated, chromatin more relaxed, and interaction with P-TEFb and BRD4 established via superenhancer contact, culminating in Pol II pause-release and productive transcription. Many other interacting proteins support these chromatin and spatial changes. While functionally the importance of AIRE binding to inactive chromatin is more obvious, the rationale of inter-action with active genes has yet to be explored. One option may be to increase